lynkr 7.2.4 → 8.0.0

package/src/config/index.js

@@ -62,7 +62,7 @@ function resolveConfigPath(targetPath) {
   return path.resolve(normalised);
 }
 
-const SUPPORTED_MODEL_PROVIDERS = new Set(["databricks", "azure-anthropic", "ollama", "openrouter", "azure-openai", "openai", "llamacpp", "lmstudio", "bedrock", "zai", "vertex"]);
+const SUPPORTED_MODEL_PROVIDERS = new Set(["databricks", "azure-anthropic", "ollama", "openrouter", "azure-openai", "openai", "llamacpp", "lmstudio", "bedrock", "zai", "vertex", "moonshot"]);
 const rawModelProvider = (process.env.MODEL_PROVIDER ?? "databricks").toLowerCase();
 
 // Validate MODEL_PROVIDER early with a clear error message
@@ -132,16 +132,24 @@ const zaiApiKey = process.env.ZAI_API_KEY?.trim() || null;
 const zaiEndpoint = process.env.ZAI_ENDPOINT?.trim() || "https://api.z.ai/api/anthropic/v1/messages";
 const zaiModel = process.env.ZAI_MODEL?.trim() || "GLM-4.7";
 
+// Moonshot AI (Kimi) configuration - OpenAI-compatible API
+const moonshotApiKey = process.env.MOONSHOT_API_KEY?.trim() || null;
+const moonshotEndpoint = process.env.MOONSHOT_ENDPOINT?.trim() || "https://api.moonshot.ai/v1/chat/completions";
+const moonshotModel = process.env.MOONSHOT_MODEL?.trim() || "kimi-k2-turbo-preview";
+
 // Vertex AI (Google Gemini) configuration
 const vertexApiKey = process.env.VERTEX_API_KEY?.trim() || process.env.GOOGLE_API_KEY?.trim() || null;
 const vertexModel = process.env.VERTEX_MODEL?.trim() || "gemini-2.0-flash";
 
+// Suggestion mode model override
+// Values: "default" (use MODEL_DEFAULT), "none" (skip LLM call), or a model name
+const suggestionModeModel = (process.env.SUGGESTION_MODE_MODEL ?? "default").trim();
+
 // Hot reload configuration
 const hotReloadEnabled = process.env.HOT_RELOAD_ENABLED !== "false"; // default true
 const hotReloadDebounceMs = Number.parseInt(process.env.HOT_RELOAD_DEBOUNCE_MS ?? "1000", 10);
 
-// Hybrid routing configuration
-const preferOllama = process.env.PREFER_OLLAMA === "true";
+// Routing configuration
 const fallbackEnabled = process.env.FALLBACK_ENABLED !== "false"; // default true
 const ollamaMaxToolsForRouting = Number.parseInt(
   process.env.OLLAMA_MAX_TOOLS_FOR_ROUTING ?? "3",
@@ -200,6 +208,12 @@ const tokenBudgetWarning = Number.parseInt(process.env.TOKEN_BUDGET_WARNING ?? "
 const tokenBudgetMax = Number.parseInt(process.env.TOKEN_BUDGET_MAX ?? "180000", 10);
 const tokenBudgetEnforcement = process.env.TOKEN_BUDGET_ENFORCEMENT !== "false"; // default true
 
+// TOON payload compression (opt-in)
+const toonEnabled = process.env.TOON_ENABLED === "true"; // default false
+const toonMinBytes = Number.parseInt(process.env.TOON_MIN_BYTES ?? "4096", 10);
+const toonFailOpen = process.env.TOON_FAIL_OPEN !== "false"; // default true
+const toonLogStats = process.env.TOON_LOG_STATS !== "false"; // default true
+
 // Smart tool selection configuration (always enabled)
 const smartToolSelectionMode = (process.env.SMART_TOOL_SELECTION_MODE ?? "heuristic").toLowerCase();
 const smartToolSelectionTokenBudget = Number.parseInt(
@@ -301,37 +315,39 @@ if (modelProvider === "bedrock" && !bedrockApiKey) {
   );
 }
 
-// Validate hybrid routing configuration
-if (preferOllama) {
-  if (!ollamaEndpoint) {
-    throw new Error("PREFER_OLLAMA is set but OLLAMA_ENDPOINT is not configured");
-  }
-  if (fallbackEnabled && !SUPPORTED_MODEL_PROVIDERS.has(fallbackProvider)) {
-    throw new Error(
-      `FALLBACK_PROVIDER must be one of: ${Array.from(SUPPORTED_MODEL_PROVIDERS).join(", ")}`
-    );
-  }
+// Deprecation warning for PREFER_OLLAMA
+if (process.env.PREFER_OLLAMA) {
+  console.warn('[DEPRECATION] PREFER_OLLAMA is removed. Use TIER_* env vars for routing. See documentation/routing.md');
+}
 
-  // Prevent local providers from being used as fallback (they can fail just like Ollama)
+// Warn about misconfigured fallback provider (only when tier routing is active,
+// since that's the only path that triggers provider fallback)
+const tiersConfigured = !!(
+  process.env.TIER_SIMPLE?.trim() &&
+  process.env.TIER_MEDIUM?.trim() &&
+  process.env.TIER_COMPLEX?.trim() &&
+  process.env.TIER_REASONING?.trim()
+);
+if (fallbackEnabled && tiersConfigured) {
   const localProviders = ["ollama", "llamacpp", "lmstudio"];
-  if (fallbackEnabled && localProviders.includes(fallbackProvider)) {
+  if (localProviders.includes(fallbackProvider)) {
     throw new Error(`FALLBACK_PROVIDER cannot be '${fallbackProvider}' (local providers should not be fallbacks). Use cloud providers: databricks, azure-anthropic, azure-openai, openrouter, openai, bedrock`);
   }
-
-  // Ensure fallback provider is properly configured (only if fallback is enabled)
-  if (fallbackEnabled) {
-    if (fallbackProvider === "databricks" && (!rawBaseUrl || !apiKey)) {
-      throw new Error("FALLBACK_PROVIDER is set to 'databricks' but DATABRICKS_API_BASE and DATABRICKS_API_KEY are not configured. Please set these environment variables or choose a different fallback provider.");
-    }
-    if (fallbackProvider === "azure-anthropic" && (!azureAnthropicEndpoint || !azureAnthropicApiKey)) {
-      throw new Error("FALLBACK_PROVIDER is set to 'azure-anthropic' but AZURE_ANTHROPIC_ENDPOINT and AZURE_ANTHROPIC_API_KEY are not configured. Please set these environment variables or choose a different fallback provider.");
-    }
-    if (fallbackProvider === "azure-openai" && (!azureOpenAIEndpoint || !azureOpenAIApiKey)) {
-      throw new Error("FALLBACK_PROVIDER is set to 'azure-openai' but AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_API_KEY are not configured. Please set these environment variables or choose a different fallback provider.");
-    }
-    if (fallbackProvider === "bedrock" && !bedrockApiKey) {
-      throw new Error("FALLBACK_PROVIDER is set to 'bedrock' but AWS_BEDROCK_API_KEY is not configured. Please set this environment variable or choose a different fallback provider.");
-    }
+  let fallbackMisconfigured = false;
+  if (fallbackProvider === "databricks" && (!rawBaseUrl || !apiKey)) {
+    fallbackMisconfigured = true;
+  }
+  if (fallbackProvider === "azure-anthropic" && (!azureAnthropicEndpoint || !azureAnthropicApiKey)) {
+    fallbackMisconfigured = true;
+  }
+  if (fallbackProvider === "azure-openai" && (!azureOpenAIEndpoint || !azureOpenAIApiKey)) {
+    fallbackMisconfigured = true;
+  }
+  if (fallbackProvider === "bedrock" && !bedrockApiKey) {
+    fallbackMisconfigured = true;
+  }
+  if (fallbackMisconfigured) {
+    console.warn(`[WARN] FALLBACK_PROVIDER='${fallbackProvider}' is enabled but missing credentials. Fallback will not work until configured.`);
   }
 }
 
@@ -589,6 +605,11 @@ var config = {
     apiKey: vertexApiKey,
     model: vertexModel,
   },
+  moonshot: {
+    apiKey: moonshotApiKey,
+    endpoint: moonshotEndpoint,
+    model: moonshotModel,
+  },
   hotReload: {
     enabled: hotReloadEnabled,
     debounceMs: Number.isNaN(hotReloadDebounceMs) ? 1000 : hotReloadDebounceMs,
@@ -596,8 +617,7 @@ var config = {
   modelProvider: {
     type: modelProvider,
     defaultModel,
-    // Hybrid routing settings
-    preferOllama,
+    suggestionModeModel,
     fallbackEnabled,
     ollamaMaxToolsForRouting,
     openRouterMaxToolsForRouting,
@@ -615,6 +635,13 @@ var config = {
   },
   logger: {
     level: process.env.LOG_LEVEL ?? "info",
+    file: {
+      enabled: process.env.LOG_FILE_ENABLED === "true",
+      path: process.env.LOG_FILE_PATH ?? path.join(process.cwd(), "logs", "lynkr.log"),
+      level: process.env.LOG_FILE_LEVEL ?? "debug",      // File captures everything
+      frequency: process.env.LOG_FILE_FREQUENCY ?? "daily", // daily | hourly | <milliseconds>
+      maxFiles: parseInt(process.env.LOG_FILE_MAX_FILES ?? "14", 10),
+    },
   },
   sessionStore: {
     dbPath: sessionDbPath,
@@ -699,8 +726,8 @@ var config = {
   semanticCache: {
     enabled: process.env.SEMANTIC_CACHE_ENABLED !== 'false',  // Disable via env if needed
     similarityThreshold: parseFloat(process.env.SEMANTIC_CACHE_THRESHOLD || '0.95'),  // Higher threshold
-    maxEntries: 500,
-    ttlMs: 3600000,  // 1 hour
+    maxEntries: Number.parseInt(process.env.SEMANTIC_CACHE_MAX_ENTRIES ?? "50", 10),  // Reduced from 500 to prevent memory bloat
+    ttlMs: Number.parseInt(process.env.SEMANTIC_CACHE_TTL_MS ?? "300000", 10),  // 5 minutes (was 1 hour)
   },
   agents: {
     enabled: agentsEnabled,
@@ -760,6 +787,12 @@ var config = {
     max: tokenBudgetMax,
     enforcement: tokenBudgetEnforcement,
   },
+  toon: {
+    enabled: toonEnabled,
+    minBytes: Number.isNaN(toonMinBytes) ? 4096 : toonMinBytes,
+    failOpen: toonFailOpen,
+    logStats: toonLogStats,
+  },
   smartToolSelection: {
     enabled: true,  // HARDCODED - always enabled
     mode: smartToolSelectionMode,
@@ -852,6 +885,23 @@ var config = {
     taskTimeoutMs: Number.isNaN(workerTaskTimeoutMs) ? 5000 : workerTaskTimeoutMs,
     offloadThresholdBytes: Number.isNaN(workerOffloadThresholdBytes) ? 10000 : workerOffloadThresholdBytes,
   },
+
+  // Intelligent Routing
+  routing: {
+    weightedScoring: true,
+    costOptimization: true,
+    agenticDetection: true,
+  },
+
+  // Model Tier Configuration (REQUIRED)
+  // Format: TIER_<LEVEL>=provider:model (e.g., TIER_SIMPLE=ollama:llama3.2)
+  modelTiers: {
+    enabled: true,
+    SIMPLE: process.env.TIER_SIMPLE?.trim() || null,
+    MEDIUM: process.env.TIER_MEDIUM?.trim() || null,
+    COMPLEX: process.env.TIER_COMPLEX?.trim() || null,
+    REASONING: process.env.TIER_REASONING?.trim() || null,
+  },
 };
 
 /**
@@ -876,15 +926,23 @@ function reloadConfig() {
   config.zai.model = process.env.ZAI_MODEL?.trim() || "GLM-4.7";
   config.vertex.apiKey = process.env.VERTEX_API_KEY?.trim() || process.env.GOOGLE_API_KEY?.trim() || null;
   config.vertex.model = process.env.VERTEX_MODEL?.trim() || "gemini-2.0-flash";
+  config.moonshot.apiKey = process.env.MOONSHOT_API_KEY?.trim() || null;
+  config.moonshot.model = process.env.MOONSHOT_MODEL?.trim() || "kimi-k2-turbo-preview";
 
   // Model provider settings
   const newProvider = (process.env.MODEL_PROVIDER ?? "databricks").toLowerCase();
   if (SUPPORTED_MODEL_PROVIDERS.has(newProvider)) {
     config.modelProvider.type = newProvider;
   }
-  config.modelProvider.preferOllama = process.env.PREFER_OLLAMA === "true";
   config.modelProvider.fallbackEnabled = process.env.FALLBACK_ENABLED !== "false";
   config.modelProvider.fallbackProvider = (process.env.FALLBACK_PROVIDER ?? "databricks").toLowerCase();
+  config.modelProvider.suggestionModeModel = (process.env.SUGGESTION_MODE_MODEL ?? "default").trim();
+
+  config.toon.enabled = process.env.TOON_ENABLED === "true";
+  const newToonMinBytes = Number.parseInt(process.env.TOON_MIN_BYTES ?? "4096", 10);
+  config.toon.minBytes = Number.isNaN(newToonMinBytes) ? 4096 : newToonMinBytes;
+  config.toon.failOpen = process.env.TOON_FAIL_OPEN !== "false";
+  config.toon.logStats = process.env.TOON_LOG_STATS !== "false";
 
   // Log level
   config.logger.level = process.env.LOG_LEVEL ?? "info";
@@ -896,4 +954,30 @@ function reloadConfig() {
 // Make config mutable for hot reload
 config.reloadConfig = reloadConfig;
 
+/**
+ * Check if any TIER_* value references Ollama (starts with "ollama:")
+ * Used by server.js to decide whether to wait for Ollama at startup.
+ */
+config.tiersReferenceOllama = function tiersReferenceOllama() {
+  const tiers = config.modelTiers;
+  if (!tiers?.enabled) return false;
+  return [tiers.SIMPLE, tiers.MEDIUM, tiers.COMPLEX, tiers.REASONING]
+    .some(v => typeof v === 'string' && v.startsWith('ollama:'));
+};
+
+// Validate TIER_* configuration (warn if missing, don't crash)
+const missingTiers = [];
+if (!config.modelTiers.SIMPLE) missingTiers.push('TIER_SIMPLE');
+if (!config.modelTiers.MEDIUM) missingTiers.push('TIER_MEDIUM');
+if (!config.modelTiers.COMPLEX) missingTiers.push('TIER_COMPLEX');
+if (!config.modelTiers.REASONING) missingTiers.push('TIER_REASONING');
+
+if (missingTiers.length > 0) {
+  config.modelTiers.enabled = false;
+  console.warn(
+    `[WARN] Missing tier configuration: ${missingTiers.join(', ')} — tiered routing disabled.\n` +
+    `  Set TIER_<LEVEL>=provider:model to enable (e.g., TIER_SIMPLE=ollama:llama3.2)`
+  );
+}
+
 module.exports = config;

package/src/context/toon.js

@@ -0,0 +1,173 @@
+const logger = require("../logger");
+
+let cachedEncode;
+let cachedLoadError;
+let warnedMissingDependency = false;
+
+function normaliseSettings(settings = {}) {
+  const minBytesRaw =
+    typeof settings.minBytes === "number" ? settings.minBytes : Number.parseInt(settings.minBytes ?? "4096", 10);
+  return {
+    enabled: settings.enabled === true,
+    minBytes: Number.isFinite(minBytesRaw) && minBytesRaw > 0 ? minBytesRaw : 4096,
+    failOpen: settings.failOpen !== false,
+    logStats: settings.logStats !== false,
+  };
+}
+
+function resolveEncodeFn(overrideEncode) {
+  if (typeof overrideEncode === "function") return overrideEncode;
+  if (cachedEncode !== undefined) return cachedEncode;
+  try {
+    const toon = require("@toon-format/toon");
+    cachedEncode = typeof toon?.encode === "function" ? toon.encode : null;
+    cachedLoadError = cachedEncode ? null : new Error("Missing encode() export from @toon-format/toon");
+  } catch (err) {
+    cachedEncode = null;
+    cachedLoadError = err;
+  }
+  return cachedEncode;
+}
+
+function looksLikeJsonObjectOrArray(text) {
+  if (typeof text !== "string") return false;
+  const trimmed = text.trim();
+  if (trimmed.length < 2) return false;
+  return (
+    (trimmed.startsWith("{") && trimmed.endsWith("}")) ||
+    (trimmed.startsWith("[") && trimmed.endsWith("]"))
+  );
+}
+
+function safeJsonParse(text) {
+  try {
+    return JSON.parse(text);
+  } catch {
+    return null;
+  }
+}
+
+function toToonString(encodeFn, value) {
+  const encoded = encodeFn(value);
+  if (typeof encoded === "string") return encoded;
+  if (encoded && typeof encoded[Symbol.iterator] === "function") {
+    return Array.from(encoded).join("\n");
+  }
+  return "";
+}
+
+function compressStringContent(content, cfg, encodeFn, stats) {
+  if (typeof content !== "string") return content;
+
+  const originalBytes = Buffer.byteLength(content, "utf8");
+  if (originalBytes < cfg.minBytes) {
+    stats.skippedBySize += 1;
+    return content;
+  }
+
+  stats.candidateCount += 1;
+  if (!looksLikeJsonObjectOrArray(content)) {
+    stats.skippedByShape += 1;
+    return content;
+  }
+
+  const parsed = safeJsonParse(content);
+  if (!parsed || typeof parsed !== "object") {
+    stats.skippedByParse += 1;
+    return content;
+  }
+
+  const toonText = toToonString(encodeFn, parsed);
+  if (typeof toonText !== "string" || toonText.trim().length === 0) {
+    return content;
+  }
+
+  const compressedBytes = Buffer.byteLength(toonText, "utf8");
+  stats.convertedCount += 1;
+  stats.originalBytes += originalBytes;
+  stats.compressedBytes += compressedBytes;
+  return toonText;
+}
+
+function applyToonCompression(payload, settings = {}, options = {}) {
+  const cfg = normaliseSettings(settings);
+  const stats = {
+    enabled: cfg.enabled,
+    available: true,
+    convertedCount: 0,
+    candidateCount: 0,
+    skippedBySize: 0,
+    skippedByShape: 0,
+    skippedByParse: 0,
+    failureCount: 0,
+    originalBytes: 0,
+    compressedBytes: 0,
+  };
+
+  if (!cfg.enabled) return { payload, stats };
+  if (!payload || !Array.isArray(payload.messages) || payload.messages.length === 0) {
+    return { payload, stats };
+  }
+
+  const encodeFn = resolveEncodeFn(options.encode);
+  if (typeof encodeFn !== "function") {
+    stats.available = false;
+    const err = cachedLoadError ?? new Error("TOON encoder unavailable");
+    if (!cfg.failOpen) throw err;
+    if (!warnedMissingDependency) {
+      logger.warn(
+        { error: err.message },
+        "TOON enabled but encoder dependency is unavailable; falling back to JSON",
+      );
+      warnedMissingDependency = true;
+    }
+    return { payload, stats };
+  }
+
+  for (const message of payload.messages) {
+    if (!message || typeof message !== "object") continue;
+    if (message.role === "tool") continue; // Never mutate machine-executed protocol payloads
+    try {
+      if (typeof message.content === "string") {
+        message.content = compressStringContent(message.content, cfg, encodeFn, stats);
+        continue;
+      }
+
+      if (!Array.isArray(message.content)) continue;
+      for (const block of message.content) {
+        if (!block || typeof block !== "object") continue;
+
+        // Keep protocol blocks untouched. Only compress user-language text fields.
+        if (block.type === "text" && typeof block.text === "string") {
+          block.text = compressStringContent(block.text, cfg, encodeFn, stats);
+          continue;
+        }
+
+        if (block.type === "input_text" && typeof block.input_text === "string") {
+          block.input_text = compressStringContent(block.input_text, cfg, encodeFn, stats);
+        }
+      }
+    } catch (err) {
+      stats.failureCount += 1;
+      if (!cfg.failOpen) throw err;
+    }
+  }
+
+  if (cfg.logStats && stats.convertedCount > 0) {
+    logger.info(
+      {
+        convertedCount: stats.convertedCount,
+        candidateCount: stats.candidateCount,
+        originalBytes: stats.originalBytes,
+        compressedBytes: stats.compressedBytes,
+      },
+      "TOON compression applied to message context",
+    );
+  }
+
+  return { payload, stats };
+}
+
+module.exports = {
+  applyToonCompression,
+};

package/src/headroom/launcher.js

@@ -5,12 +5,17 @@
  * Provides automatic container creation, health checking, and graceful shutdown.
  */
 
-const Docker = require("dockerode");
+let Docker;
+try {
+  Docker = require("dockerode");
+} catch {
+  Docker = null;
+}
 const logger = require("../logger");
 const config = require("../config");
 
-// Initialize Docker client
-const docker = new Docker();
+// Initialize Docker client (only if dockerode is available)
+const docker = Docker ? new Docker() : null;
 
 // Launcher state
 let containerInstance = null;

package/src/logger/index.js

@@ -1,4 +1,6 @@
 const pino = require("pino");
+const fs = require("fs");
+const path = require("path");
 const config = require("../config");
 const { createOversizedErrorStream } = require("./oversized-error-stream");
 
@@ -64,6 +66,27 @@ streams.push({
 			: process.stdout,
 });
 
+// File rotation stream (if enabled via LOG_FILE_ENABLED=true)
+if (config.logger.file?.enabled) {
+	const fileConfig = config.logger.file;
+	// Ensure log directory exists
+	const logDir = path.dirname(fileConfig.path);
+	fs.mkdirSync(logDir, { recursive: true });
+
+	streams.push({
+		level: fileConfig.level,
+		stream: pino.transport({
+			target: "pino-roll",
+			options: {
+				file: fileConfig.path,
+				frequency: fileConfig.frequency,
+				limit: { count: fileConfig.maxFiles },
+				mkdir: true,
+			},
+		}),
+	});
+}
+
 // Oversized error stream (if enabled)
 if (config.oversizedErrorLogging?.enabled) {
 	streams.push({